Abstract:
Antimicrobial resistance (AMR) poses a significant global threat to public health. In
recent years, machine learning (ML) algorithms have emerged as tools for predicting
AMR patterns and guiding antibiotic treatment decisions. This study aimed to explore
the predictive capabilities of three ML algorithms, namely Extreme Gradient
Boosting (XGBoost), Support Vector Machines (SVM), and Random Forest (RF),
with the incorporation of integron features along with AMR genes, to construct a
web-based in silico antibiogram tool. Utilizing a comprehensive dataset comprising
bacterial isolates - Acinetobacter baumannii, Escherichia coli, and Klebsiella pneumoniae
and their corresponding resistance profiles to five antibiotics of interest -
cefotaxime, ceftriaxone, ciprofloxacin, gentamicin, and levofloxacin, the study train
and evaluate the performance of the ML models. Generally, XGBoost and RF perform
better than SVM with AUC score up to 0.93. Central to this approach is the
integration of integron features, which play a pivotal role in mediating horizontal gene
transfer and facilitating the dissemination of resistance genes among bacterial populations.
Building upon the robust models, this study developed a user-friendly web application
that enables healthcare practitioners and researchers to input whole-genome
sequences and rapidly obtain predictions regarding antibiotic resistance profiles based
on integrated ML models.
